Towards accurate non-contact moisture inspection using THz imaging and thickness information

Abstract This study proposes an approach for a non-contact moisture inspection in dried food products, which is crucial to maintain optimal quality and shelf-life, using terahertz (THz) signal. To achieve this, a sample-specific calibration curve needs to be determined first. HAITAI crackers were chosen in this work for demonstration purposes. Fifteen stacks of crackers with different heights were prepared and moisturized by covering with a wet tissue paper for different time periods, resulting in moisture levels between 3 and 40% R.H.. Then, each sample was placed on a conveyor belt system between a THz source and THz a detector, and transmitted signal was measured 5 times. After that, moisture percentage of the sample was determined based on a gravimetric method, whose results served as a ground-truth measurement. A thickness of the sample was also measured using a vernier. All signal measurements, together with their corresponding known thicknesses and moisture percentages, were used to calculate necessary coefficients that define a sample-specific calibration curve. Once a calibration curve for the cracker was obtained, it was used to estimate the moisture percentages in samples with different thicknesses. Mean absolute error (MAE) of the moisture percentage is found to be less than 12% when the sample thickness is modelled as part of the calibration curve, which is over 50 times less compared to the case when the sample thickness is not modelled. Therefore, the utilization of an automatic thickness determination would be promising for real-time and accurate non-contact moisture inspection. This approach can be also integrated into a production line to improve quality control in the food industry without interrupting existing processes.